视频选题榜 2026-09-11

• https://arxiv.org/abs/2609.09264 · StochBench: A Domain-Specific Benchmark for Stochastic Processes in Lean · 研究生级随机过程首个 Lean 评测集,Opus 4.8 仅 34.9% 证明率,揭穿聚合分骗术 · round=R1
• https://arxiv.org/abs/2609.10266 · KVShareArena: KV-Cache Reuse across Contexts and Model Checkpoints · 首个非前缀复用基准,位置修正救单源,多源问题只有部分重计算恢复 50–67% gap · round=R2
• https://arxiv.org/abs/2609.11596 · From Intent to Execution Grant: An Execution-Boundary Conformance Profile for High-Risk AI Actions · 执行边界契约级判定,5 类外部后果动作×4 工程原语交集,ERC 非授权令牌 + Redemption-time 二次校验 · round=R3